CA1258296A - Decision feedback equalizer with a pattern detector - Google Patents

Abstract

ABSTRACT OF THE DISCLOSURE
A decision feedback equalizer has an adaptive filter responsive to a demodulated data sequence and a residual intersymbol interference (ISI) signal for estimating ISI which occurs during pulse transmission of a period of T seconds and producing an estimated ISI signal. A subtractor subtracts the estimated ISI signal from a received signal which involves an ISI signal to produce a residual signal. A circuit is provided for extracting the residual ISI signal from the residual signal and a delayed residual signal which is produced by delaying the residual signal. A demodulator produces the demodulated data sequence from the residual signal and applies the data sequence to the filter. A pattern detector detects a particular consecutive pattern out of the demodulated data sequence. The residual ISI signal is applied to the filter in response to an output of the pattern detector. The invention dispenses with the need for an automatic gain control circuit, thereby permitting scaling down the hardware.

Description

DECISION FEEDBACK E~UALIZER
r~ITH A PATTERN DETECTOR

Background of the Invention The present invention relates to a decision feedback equalizer (DFE) for removing intersymbol interference (ISI) which is particular to pulse transmission.
A DFE is one of implementations heretofore proposed for the removal of ISI which occurs during pulse transmission. In a DFE, an adaptive filter having taps corresponding to a length over which ISI has influence is installed to generate estimated ISI, so that ISI
occurring during transmission of pulses over a channel may be suppressed. The tap coefficients of the filter are sequentially corrected by determining a correlationship between residual ISI and a result of decision of a received signal.
A problem with a DFE is that the adaptive operation is unachievable unless residual ISI contained in a residual signal which is produced by subtracting estimated ISI from a received signal having ISI, is detected with accuracy in the event of correction of the coefficients.
For example, where a biphase code or like two-level code which will be described later is used as a transmission line code, extracting ISI only is impractical because the received signal level has no zero level duration due '~

~:~5~

2 --to -the particular nature of two-level codes, the DFE thus failing to correct the tap coefficients.
A solution to the above problem is proposed in a paper entitled "Some Considerations on the Design of Adaptive Digital Filters Equipped with the Sign Algorithm", IEEE Transactions on Communications, Vol. COM-32, No. 3, - March 1984, pp. 258-266. The approach as described in this paper is equipping a DFE with a subtractor and an automatic gain control (AC-C) circuit so as to remove all signals except ISI. Such a scheme, however, undesirably scales up hardware partly because the AGC circuit is essential and partly because a complicated control is necessary for a signal which is fed from the AGC to the subtractor and free from ISI to be maintained at an adequate level.

Summary of the Invention It is therefore an object of the present invention to eliminate the drawback particular to the prior art DFE and provide a DFE which is simple in construction and small in scale.
, A DFE in accordance with the present invention comprises first adaptive filter means responsive to a demodulated data sequence and a residual ISI signal for estimating ISI which occurs during pulse transmission f a period of T second and producing an estimated ISI

~5~316

-3- 6446-407 signal, first subtractor means for subtracting the estimated ISI
signal from a received signal having an ISI signal to produce a residual signal, means for extracting the residual ISI signal from the residual signal and a delayed residual signal which is pro-duced by delaying the residual signal, first demodulating means for producing the demodulated da-ta sequence from the residual sig-nal and applying the demodulated data sequence to the first adap-tive filter means, pattern detector means for detecting a parti-cular consecutive pattern out of the demodulated data sequence, and means for applying the residual ISI signal to the filter means in response to an output of the pattern detector means.
Brief Description of the Drawings The above and other objects, features and advantages of the present invention will become more apparent from the follow-ing detailed description taken with the accompanying drawings in which:
Figs. 1 and 2 show waveforms representa-tive of trans mission channel codes;
Fig. 3 is a block diagram of a DFE with a pattern detec-tor embodying the present invention;
Figs. 4A to 4D and Figs. 6A to 6F show waveforms forexplaining the principle of the pattern detector;
Fig. 5 is a circuit diagram of a first embodiment of pattern detector;
Fig. 7 is a circuit diagram of a second embodiment of pattern detector;

:

~5~
-3a- 6446-407 Figs. 8, 11, 13 and 14 are circuit diagrams each show-ing a specific construction of an adaptive filter;

~L25~

Fiy. 9 is a diagram of a coefficient generator which is included in the filteri Fig. 10 is a block cliagram showing a second embodiment of the present invention; and Fig. 12 is a block diagram showing a third embodiment of the present invention.
It is -to be noted that throughout the drawings the same or similar structural elements are designated by like reference numerals.

Detailed Description of the Preferred Embodiments For a better understanding of the present invention, transmission line codes which are applicable to various embodiments of the present invention will be described.
Fig. l shows a biphase code and Fig. 2 a minimum shift keying (MSK) code. As shown in Fig. 1, the biphase code assigns pulse shapes of opposite polarities to data ONE and ZERO. The two pulse shapes individually have the polarities inverting at the center of one bit width~
i.e., T second and are balanced within one bit in terms of positive and negative polarities. In contrast, the MSK code is implemented with four different kinds of pulse shapes, as shown in Fig. 2. Specifically, the MSK
code assigns two different kinds of pulse shapes of opposite polarities, i.e., a ONE mode and a ZERO mode ~5~32~6 to each of data ZERO and ONE, the transitions of the four pulse shapes being represented by arrows in Fig. 2. The MSK code is characterized in that the polarity is necessarily inverted at the point of interconnection of transmitted pulse shapes. As shown in Fig. 2, the MSK
code is balanced for a data ONE within one pulse shape but not for a data ZERO, each in terms of positive-negative levels. Nevertheless, it will be seen from the directions of the arrows of Fig. 2 that if an even number of data ZEROs exist in a data seguence, the positive~
negative balance is set up so that the D~ component is almost negligible.
Referring to Fig. 3, a DFE in accordance with the present invention is shown and includes an input terminal 1.
A received signal which involves ISI is applied to the input terminal 1 and then to a subtractor 2. The subtractor 2 subtracts from the received signal an estimated ISI
signal which is generated by an adaptive filter 5, -thereby producing a residual signal which involves residual ISI.
The residual signal is applied to an arithmetic circuit 9 after being delayed by a delay element 8 by 1 sampling times. The arithmetic circuit 9 cancels the data involved in the residual signal by the delayed signal to extract a residual ISI signal. In this particular embodiment, the cancellation by the circuit 9 is effected by subtraction of the two input signals. The residual signal from the ~5~29~

subtractor 2 is also fed to a decision circuit 3 which compares the residual signal with a reference value to determine an original two-level data sequence~ This data sequence is routed to an output terminal 4, a pattern detector 11, and the adaptive filter 5. The dicision circuit 3 may be conveniently implemented with a construction which is described in a paper "A Study on the Subscriber Loop Transmission System for ISDN sased on the Echo Cancellation Technique" presented at IEEE
International Conference on Communications which was held in Chicayo, U.S.A on July 23 to 26, 1985. The pattern detector 11 detects out of the output of the decision circuit 3 a particular pattern from which the residual ISI signal can be detected as will be described later, the detected pattern being applied to a selector 10.
In response to the pattern detection signal, the selector 10 selects ei-ther one of the residual ISI signal or a zero volt and delivers it to the filter 5. Based on the output of the selector 10 and the data sequence of the decision circuit 3, the filter 5 corrects tap coefficients or filter coefficients and generates an estimated ISI
signal. It is to be noted that -the correction of filter coefficients does not occur when the selector 10 has selected a zero volt.
The reason why the residual ISI signal is obtainable by cancelling the data involved in the residual signal is that the ISI signal of the residual signal and that of the delayed signal are not correlated. Because the value of the current ISI signal and that of the delayed ISI
signal are not correlated, the value of the delayed ISI
signal may be regarded as random noise. The delayed ISI
signal has a symmetrical amplitude distribution with respect to the positive and negative polarities, and -the probability that its amplitude d becomes ¦d¦~ ~(where 0 ~ ~`) is not zero and assumes a certain positive value.
It follows that the probability that the output of the arithmetic circuit 9 includes a residual ISI signal assumes a given positive value. Further, the magnitude of the residual ISI signal is generally sufficiently smaller than a received signal.
The pattern detector 11 is constructed and operated as follows. As previously stated, to extract the residual ISI signal, it is necessary for the data involved therein to be canceled from the residual signal. Assume that the MSK code as shown in Fig. 2 is used as a transmission line code, and that the delay time iT is 2T. Then, in order to cancel the data involved in the residual signal, the pulse shape at a time t = 0 and that at a time t = -2T have to be the same. Meanwhile, because the pulse shapes at t = 0 and t = -2T need be smoothly interconnected at t = -T by the transition of the MSK
code, it is only when the code has consecutive patterns as shown in Figs. 4A ("000") and 4B ("111") that the data can be canceled. It is noted that when the line code has patterns as shown in Figs. 4C and 4D, cancellation of the data can be effected by adding the pulse shape at T = 0 to the pulse shape at t = -2T. The pattern detector 11 is constructed to detect the code sequences "000" (Fig. 4A) and "111" (Fig. 4s) as mentioned above.
Fig. 5 shows a specific construction of the pattern detector 11. In Fig. 5, the input signal 51 corresponds to the data sequence outputted by the decision circuit 3 as shown in Fig. 3, and the input signal 52 to the mode signal. While in Fig. 3 the signal path between the decision circuit 3 and the pattern detector 11 and that between the circuit 3 and the filter 5 are each represented by a single path, in the case of MSK code they are representative of two signal paths each, one for the data signal and the other for the mode signal. Delay elements 53 and 54 each adapted to provide a delay of T second and an Exclusive-NOR (EXNOR) gate 55 cooperate to determine whether the current dat~ and the data appeared 2T seconds before are coincident with each other.
Likewise, delay elements 56 and57 each adapted to provide a delay of T second and an EXNOR gate 58 cooperate to see if the current mode signal and the mode signal delayed by 2T second are coincident with each other. An AND
gate 59 provides AND of the outputs of the EXNOR gates 55 ~25;~2~
g and 58 to produce a pa-ttern detection signal 60.
The pattern detector 11 will be described in relation to the biphase code as shown ln Fig. 1 which is another possible transmission line code.
Assuming that the delay is 2T, the signals at t = 0 and t = -2T need to have the same pulse shapes in order that residual ISI may be extracted, as has been the case with the MSK co~e. However, once the code sequence "000"
as shown in Fig. 6A is deteriorated on -the transmission channel as shown in Fig. 6B, the latter half X of the pulse shape at t = 0 and the former half Y of the pulse shape at t = -2T are respectively dependent on pulse shapes at t = T and t = -3T. Specifically, if a data ONE appears at t = T, the pulse shape at t = 0 is decided as shown in Fiy. 6E and, if a data ZERO appears a-t t = T, it is decided as shown in Fig. 6F. Likewise, the pulse shape at t = -2T is decided as shown in Fig. 6C if a data ONE appears at t = 3T and as shown in Fig. 6D if a data ZERO appears thereat. It will thus be seen that while the pulse shapes of Figs. 6D and 6F are identical with each other and, therefore, can cancel each other, the pulse shape of Fig. 6C and that of Fig. 6E or that of Fig. 6F are not identical and, therefore, cannot cancel each other.
In the same manner, in the case of a waveform representative of a code sequence "111", a data ONE

~2~i~2~

appears at t = T and t = -3T; in the case of waveforms representative of "010" and "101", a data ZERO appears at t = T and t = -3T. In short, in order that a three-bits code which can be canceled may be detected, five-bit code checkiny is required; five-bit codes which can be canceled include four patterns in total, i.e."00000", "10101", "11111" and "01010".
Fig. 7 shows a specific construction of the pattern detector 11 constructed to detect such five-bit patterns.
As shown, the pattern detector includes EXNORs 71, 72 and 73 which function to see coincidence of the firs-t and third bits of a five-bit code, that of the second and fourth bits, and that of the third and fiEth bits, respectively. The detector also includes an AN~ gate 74 adapted to detect coincidence of ou-tputs of the EXNORs 71 to 73.
Referring to Fig. 8, the adaptive filter 5 of Fig. 3 is shown in a detailed block diagram. In Fig. 8, the input signals 106' and 106 respectively correspond to data (ZERO or ONE) and mode (positive or negative) which are outputted by the decision circuit 3 of Fig. 3u Further, the input output signals 107 and 108 respectively correspond to output signals of the selector 10 and adaptive filter 5 of Fig. 3. The mode signal 106 is fed to a delay element 1001, a multiplier 101o, and a coefficient generator 1020. The data signal 106', on ~251~6 the other hand, is applied to a delay element 100'1 and the coefficient generator 1020. It should be born in mind that N in Fig. 8 is representative of the number of taps which is a positive integer. In the drawing, the coefficient generators 1020 to 102N are adapted to supply their associated multipliers lOlo to 101N wi-th coefficients corresponding to the respective modes in response to the mode signal 106'. The multipliers multiply outputs of their associated delay elements and the coefficients applied thereto, delivering the products to an adder 103. Summing up the products, the adder 103 delivers an est:imated ISI signal as the output signal 108. Each of the coefficient generators 1020 to 102N
corrects its coefficient in response to a residual ISI
signal which is fed as -the input signal 107 and an output signal from its associated delay element.
Referring to Fig. 9, a specific construction of the coefficient generator 1021 (1 = 0, 1, ..., N) is shown.
In Fig. 9, the input signal 200 corresponds to the input signal 106 or one of the output signals of the delay element 1001, 1002, ..., lOON of Fig. 8. Likewise, the input signal 200' corresponds to the signal 106' or one of the outputs 100'1, 100'2, ..., lOO'N of Fig. 8, and the input signal 201 to the signal 107 of Fig. 8.
Further, the output signal 209 of Fig. 9 corresponds to the output of the coeEficient generator 1021 of Flg. 8.

~S~2~6 As shown, the data signal 200' which is representative of a ZERO or a ONE is fed as a control signal to selectors 204, 205 and 208. Meanwhile, the mode signal 200 assuming a value of -~1 or -1 which is representative of a + mode or a - mode and associated with the data signal 200' is applied to one input of a multiplier 202. Applied to the other input of the multiplier202 is a residual ISI signal 201 which comprises an ISI component only. The multiplier 202 multiplies the mode signal 200 and the residual ISI
signal 201 and delivers the product to one input of an adder 203. ~elay elements 206 and 207 serve as coefficient memories which are respectively associated with a ZERO
and a ONE of the data signal 200', the elements 206 and 207 being coupled to a selector 208. The data signal 200' is also fed to the selector 208 as a control signal When the data signal 200' is a ZERO, the selector 208 selects a coefficient which is associated with a ZERO
which is the output of the delay element 206 and, when the data signal 200' is a ONE, it selects a coefficient associated with a ONE which is the output of the delay element 207. The coefficient 209 selected by the 'selector 208 is delivered as indicated as in Fig. 9.
This coefficient 209 is routed also to the adder 203 to be added to an output of the multiplier 202. The 25 output of the adder 203 is fed to the selec-tors 204 and 205. The outputs of the delay elements 206 and 207 are ~25~2~

applied also to the selectors 204 and 205, respectively.
The outputs of the selectors 204 and 205 are applied to the delay elements 206 and 207, respectively. The selectors 204, 205 and 208 operate as follows. Assume that the data signal 200' is a ZERO. Then, the selector 208 selects the output of the delay element 206 which corresponds to the data ZERO, the selected delay output being delivered as a coefficient 209. The coefficient is fed to the adder 203 and then fed back to the delay element 206 via the selector 204, thereby updating the coefficient which is associated with the data ZERO.
Meanwhile, the selector 205 selects the output of the delay element 207 and feeds it back to the delay element 207 with the result that the coefficient associated with the data ONE is not updated. Conversely, when the data signal 200' is a ONE, the selector 208 selects the output of the delay element 207 which is associated with the data ONE. The resultant coefficient 209 is applied to the adder 203 and then fed back to the delay elemen-t 207 via the selector 205, whereby the coefficient associated with the data ONE iS updated. The selector 204, on the other hand, selects the output of the delay element 206 and returns it to the element 206 with the result that the coefficient associated wi-th the data ZERO is not updated. By the principle of operation described above, a particular coefficient which is associated with the ~ 2~ 6 value of the data signal, i.e., a ZERO or a ONE is selected for the arithmetic operation of the filter and, at the same time, the coefficient used is updated while the coefficient not used is not updated. In this manner, the coefficients of the filter are set up adaptively.
Referring to Fig. 10, a second embodiment of the present lnvention is shown. As well known in the art, the zero-crossing point of a received signal changes as the transmission distance increases. Should the received signal be decided based on a clock signal which is derived from such a signal, the decision point would be deviated from a point where the data reaches the highest level.
The embodiment of Fig. 10 is constructed to eliminat:e such an occurrence.
In Fig. 10, the DFE includes a selector 13 adapted to select either the output of the selector 10 or the residual signal in response to an output of the dec:ision circuit 3, and a switch 14. ~he switch 14 has a fourth terminal 144 for receiving the residual signal, a firs-t and a third terminals 141 and 143 for receiving an output of the selector 10, and a second terminal 142 for receiving an output of the selector 13. At times tl = T/4, t2 = 2/4 T, t3 = 3/4 T and t4 = T, the switch I'L
sequentially selects the terminals 141 to 144 in response to the control signal. The signal selected by the switch 14 is fed to an adaptive filter 50~ The filt:er 50 ~5~;~9~

adjusts coefficients thereof such that the residual ISI
at each time converges to zero.
Referring to Fig. 11, a specific construction of the adaptive filter 50 is shown in a block diagram.
The filter 50 includes, in addition to the structural elements of the filter of Fig. 8, multipliers, coefficient generators and adders which serve to process signals appearing at times t2 = 2/4 ~T, t3 = 3/4 T and t4 = T.
In detail, multipliers 1011 0 to 1011 N~ coefficient generators 1021 0 to 1021 N and an adder 1030 are used for a data sequence appearing at t2 = 2/4 T, multipliers 10130 to 1013N, coefficient generators 10230 to 1023N
and an adder 1033 for a data sequence appearing at -t4 = T. The delay elements 1001 to lOON and 100'1 to lOO'N are shared by the data sequences at 2/4-Tto4/4 T.
Refèrring to Fig. 12, a third embodiment of the present invention is shown. While the first and second embodiments each contemplates to remove ISI with which the past pulse shape effects the current pulse shape, the third embodiment further contemplates to remove intrasymbol interference (ITSI) with which the current pulse shape itself effects the current levels at times 2/4 T and 4/4 T. As shown, the DFE of Fig. 12 includes a switch 16 for distributing residual signal of the subtractor 2 at times tl = T/4, t2 = 2/4 T and t4 = T.
The signal at T/4 from the switch 16 is applied as data 9~i to a filter 18 while the output of the switch 16 at t4 = T
is regarded as an ITSI signal which appears at t4 = T due to the pulse shape and is applied as a coefficient correction signal toa filter 20. Further, the output of the switch 16 at 2/4 T is regarded as an ITSI signal which appears at t = 2/4 T due to the former half of the pulse shape and is applied as a coefficient correction signal to the filter 18. A selector 17 delivers to the filter 18 the signal appearing a-t 2/4 T or a zero volt when the output of the decision circuit 3 is a ONE or a ZERO, respectively. In response to the signal at T/4 from the switch 16 and the ITSI signal at 2/4 T from the selector 17, the filter 18 generates an estimated ITSI signal for removing the ITSI signal at 2/4 T.
Likewise, the filter 20 responds to the data outputted by the de-tection circuit 19 and the ITSI signal at T
by generating an estimated ITSI signal which removes the ITSI signal at T. An adder 21 sums outputs of the filters 50 and 18 to apply the sum to an adder 22 while the adder 22 sums an output of the filter 20 and that of the adder 21 to generate an estimated ITSI slgnal.
Fig. 13 shows a specific construction of the adaptive filter 18 of Fig. 12. In Fig. I3, the input signals 300 and 301 respectively correspond to an output signal of a first terminal of the switch 16 of Fig. 12, i.e., polarity of the residual signal at the sampling time tl, ~.2~9~

and an output of the selector 17, i.e. polarity of an ITSI signal at the sampling time t2 or an error signal which is a zero volt. Further, the output signal 306 as shown ln Fig. 13 corresponds to the output signal, or estimated ITSI, of the adaptive filter 18 of Fig. 12.
In Fig. 13, the polarity 300 of the residual signal is routed to multipliers 302 and 305. A delay element 304 adapted for a delay of T second serves as coefficient memory and applies an output thereof to the multiplier 305 which then generates estimated ISI 306. Also, the output of the delay element 304 is fed back thereto via an adder 303 so that a coefficient associated with the output of the multiplier 302 is selectively updated. When the error signal 301 is zero, the output of the multiplier 302 is also zero and, therefore, the coefficient is not updated.
In this manner, the coefficient is selectively updated.
In the output of the adaptive filter 18 of Fig. 12, the estimated ITSI signal at the zero-crossing point which occurs at the center within a symbol appears and is added by the adder 21 to the estimated ITSI signal which is generated by the adaptive filter 50. The output of the adder 21 is fed to the subtractor 2 via the adder 22.
Referring to Fig. 14, a specific construction of the adaptive filter 20 of Fig. 12 is shown in a detailed block diagram. The filter 20 corresponds to only one phase and one tap of the filter 50. The principle of . ~

g~

operation of the filter 20 for updating the coefficients is the same as that of Fig. 9 and, therefore, detailed description thereof will be omitted to avoid redundancy.
In Fig. 14, the same function blocks and signals as those of Fig. 9 are designated by like reference numerals.
It should be noted, however, that in Fig. 14 the data signal 200 and the mode signal 200' correspond to the outputs of the decision clrcuit 3 of Fig. 10, and the error signal 201 to the output of the fourth terminal of the switch 1~ as shown in Fig. 12. The filter of Fig. 14 differs from that of Fig. 9 in that the mode signal 200' is multiplied by a coefficient 209 by a multiplier 210 to produce estimated ITSI 211. Another difference is that while the error signal 201 in Fig. 9 selectively assumes three values, i.e., +1 and 0, it in Fig. 14 assumes either one of two values, i.e. +1. The estimated ITSI which is outputted by the adaptive filter 20 is applied to the adder 22 to be added to an output of the adder 21, the sum being fed the subtractor 2.
In all the embodiments shown and described, the arithmetic unit 9 performs subtraction on an input signal and a delayed signal in order to cancel a data involved in a residual signal. If desired, the subtraction may be replaced with addition in which case a consecutive symbol pattern is selected such that, as shown in Figs.
4C and 4D, the wave shape of one of the current input 5 ~ d signal and the delayed signal becomes identical with pulse shape of the other when inverted. In the pattern detector 11 of Fig. 5, the EXNOR 58 responsive to coincidence of mode signals may be replaced with an Exclusive-OR gate which is responsive to non-coincidence of mode signals.
Although the sampling period T/R has been assumed to be T/4 second in the foregoing description, it will be apparent that the principle of the present invention is effective so long as R is a positive even number.
Furthermore, while the embodiments of Figs. 10 and 12 have been described in relation to MSK code of Fig. 2, the present invention is similarly applicable to biphase code as shown in Fig. 1.
In summary, it will be seen that in accordance with the present invention the adaptive operation of an adaptive filter is ensured because the filter is controlled such that the coefficients are selectively ~Ipdated by detecting the pattern of a received signal waveform appearing when either a sum or a difference between the current value of a residual signal and a value of the same appeared iT second before is equal to the ITSI signal. This realizes decision feedback type removal of ISI which eliminates the need for complicated control and can be implemented with a simple and small-scale hardware con-figuration. In addition, the present invention allows not only ISI due to a past sequence of symbols but only ISI within a sequence of data to be removed.

Claims (4)

What is claimed is:

1. A decision feedback equalizer comprising:
first adaptive filter means responsive to a demodulated data sequence and a residual intersymbol interference (ISI) signal for estimating ISI which occurs during pulse transmission of a period of T second and producing an estimated ISI signal;
first subtraction means for subtracting the estimated ISI signal from a received signal which involves an ISI
signal to produce a residual signal;
means for extracting the residual ISI signal from the residual signal and a delayed residual signal which is produced by delaying the residual signal;
first demodulating means for producing the demodulated data sequence from the residual signal and applying the data sequence to said first filter means;
pattern detector means for detecting a particular consecutive pattern out of the demodulated data sequence;

and means for applying the residual ISI signal to said first filter means in response to an output of said pattern detector means.

2. A decision feedback equalizer as claimed in claim 1, further comprising:

first sampling switch means for sampling the residual ISI signal N times at a period of T/N so as to generate N residual ISIs;
N - 1 second adaptive filter means responsive to the N - 1 residual ISIs and an output of said decision circuit for generating N - 1 second estimated ISIs; and means for applying the N - 1 estimated ISIs sequentially to said first subtractor means at a period of T/N.

3. A decision feedback equalizer as claimed in claim 2, further comprising:
second demodulating means for producing the second demodulated data sequence from the residual signal;
second sampling switch means for sampling the residual signal N times at a period of T/N to generate a first to an "N" residual signal sample values;
third adaptive filter means in response to an output of said second demodulating means for generating a first estimated intrasymbol interference (ITSI) signal; and means for adding outputs of said third and one of the first and second filter means.

4. A decision feedback equalizer as claimed in claim 3, further comprising:
fourth adaptive filter means responsive to the first and(N/2)th residual signal sample values for generating a second estimated ITSI; and adder means for adding outputs of said fourth and one of the first and second filter means.